Confidence in probability happens when a test is done a number of times (or a survey carried out with a number of respondants) and an average response from the number of tests is calculated. Then the confidence interval is calculated based upon the number of tests (respondants) and the deviation of individual respondants from the average to allow you to say that "I am 95% (1.65 standard deviations from the average) that the result +/- X.YZ% is the actual mean for all iterations of the situation tested".
What confidence gives you is a window between which you expect most tests or responses to come. With more tests or people surveyed (as long as they are properly conducted, the testers unbiased, and in surveys responders are randomly selected) the confidence will get tighter and tighter (the law of diminishing returns kicking in over time).
Assigning a 50/50 probability to God/no God is incorrect for two reasons, 1) we've no actual tests to establish probability to any degree of accuracy, and 2) indirectly (from the study of a lot of physical phenomena and systems) we have a lot of evidence which suggests that the universe does not need a god to exist or be created, thus indicating the probability to be less than 50% in favour of God. Now this does not mean that the actual odds aren't 50/50, it just means that given our current knowledge we have nothing to suggest that they are.